Method for generating a reduced order model of an electronic circuit

- Motorola Inc.

A method for model reduction (48) for electronic circuit simulation (52) of an electronic circuit using multipoint matrix Pade approximation is provided herein. Using the method, state equations are generated from a linear circuit to be analyzed. One or more expansion frequencies and a number of moments for each of the one or more expansion frequencies are provided. Starting block Lanczos vectors using a first expansion frequency of the one or more expansion frequencies and the state equations are generated. New block Lanczos vectors are generated from the starting block Lanczos vectors. The new block Lanczos vectors are scaled and normalized. Breakdowns in the new block Lanczos vectors are detected and treated to generate new starting block Lanczos vectors.

Skip to:  ·  Claims  ·  References Cited  · Patent History  ·  Patent History

Claims

1. A method for generating a reduced order model of an electronic circuit, comprising the steps of:

generating state equations from an electronic circuit to be analyzed;
providing one or more expansion frequencies and a number of moments for each of the one or more expansion frequencies;
generating a set of starting block Lanczos vectors using a first expansion frequency of the one or more expansion frequencies and the state equations;
orthogonalizing the set of starting block Lanczos vectors;
generating a new set of block Lanczos vectors from a previous set of block Lanczos vectors;
orthogonalizing the new set of block Lanczos vectors with respect to the previous set of block Lanczos vectors;
scaling and normalizing the new set of block Lanczos vectors to produce a new set of scaled and normalized block Lanczos vectors;
detecting breakdowns in the new set of scaled and normalized block Lanczos vectors, and if detected, treating the breakdowns to generate a new set of starting block Lanczos vectors from the new set of scaled and normalized block Lanczos vectors; and
determining if all moments of the first expansion frequency have been used, and if all moments have not been used, generating new sets of block Lanczos vectors from the new set of starting block Lanczos vectors, and if all moments have been used, then determining if all expansion frequencies of the one or more expansion frequencies have been used, and if all of the expansion frequencies of the one or more expansion frequencies have not been used, retrieving a next expansion frequency of the one or more expansion frequencies and generating a new starting block Lanczos vectors, and if all of the expansion frequencies have been used, providing the reduced order model of the electronic circuit.

2. The method of claim 1, further comprising the step of detecting if the reduced order model meets a predetermined accuracy by comparing state equations generated from the reduced order model to the state equations of the electronic circuit.

3. The method of claim 1, wherein a set of block Lanczos vectors is characterized as being a pair of block Lanczos vectors.

4. The method of claim 1, wherein the step of scaling and normalizing further comprises the steps of:

generating a set of orthogonal factorizations corresponding to the new set of block Lanczos vectors;
generating a singular value decomposition of the new set of block Lanczos vectors; and
scaling and normalizing the new set of block Lanczos vectors using the set of orthogonal factorizations and the singular value decomposition.

5. The method of claim 1, wherein the step of detecting breakdowns further comprises the steps of:

determining that a breakdown has occurred by checking whether the new set of scaled and normalized block Lanczos vectors has a singular matrix product;
determining that the breakdown is simple and generating a first new set of random block Lanczos vectors to become a new set of starting block Lanczos vectors;
determining that the breakdown is not simple and patching the new set of scaled and normalized block Lanczos vectors with a second new set of random block Lanczos vectors to form a new set of starting block Lanczos vectors; and orthogonalizing the new set of starting block Lanczos vectors with respect to all previously generated sets of block Lanczos vectors.

6. The method of claim 1, wherein the electronic circuit is characterized as being a linear circuit.

7. A method for model reduction for electronic circuit simulation of an electronic circuit:

generating state equations from a circuit to be analyzed;
providing one or more expansion frequencies and a number of moments for each of the one or more expansion frequencies;
generating a set of starting block Lanczos vectors using a first expansion frequency of the one or more expansion frequencies and the state equations;
generating a set of new block Lanczos vectors from the set of starting block Lanczos vectors;
scaling and normalizing the set of new block Lanczos vectors to produce a new set of scaled and normalized block Lanczos vectors; and
detecting breakdowns in the set of new block Lanczos vectors, and if detected, treating the breakdowns to generate a set of new starting block Lanczos vectors from the new set of scaled and normalized block Lanczos vectors.

8. The method of claim 7, wherein the step of scaling and normalizing further comprises the steps of:

generating a set of orthogonal factorizations corresponding to the set of new block Lanczos vectors;
generating a singular value decomposition of the set of new block Lanczos vectors; and
scaling and normalizing the set of new block Lanczos vectors using the set of orthogonal factorizations and the singular value decomposition.

9. The method of claim 8, wherein the step of detecting breakdowns further comprises the steps of:

determining that a breakdown has occurred by checking whether the new set of scaled and normalized block Lanczos vectors has a singular matrix product;
determining that the breakdown is simple and generating a first new set of random block Lanczos vectors to become a new set of starting block Lanczos vectors;
determining that the breakdown is not simple and patching the new set of scaled and normalized block Lanczos vectors with a second new set of random block Lanczos vectors to form a new set of starting block Lanczos vectors; and
orthogonalizing the new set of starting block Lanczos vectors with respect to all previously generated sets of block Lanczos vectors.

10. The method of claim 9, wherein the electronic circuit is characterized as being a linear circuit.

11. The method of claim 10, wherein a set of block Lanczos vectors is characterized as being a pair of block Lanczos vectors.

12. The method of claim 11, further comprising a step of detecting if the method for model reduction meets a predetermined accuracy by comparing state equations generated from a reduced order model to the state equations of the circuit.

13. A method for generating a multipoint matrix Pade approximation of an electronic circuit, comprising the steps of:

generating state equations from a circuit to be analyzed;
providing one or more expansion frequencies and a number of moments for each of the one or more expansion frequencies;
generating a set of starting block Lanczos vectors using a first expansion frequency of the one or more expansion frequencies and the state equations;
orthogonalizing the set of starting block Lanczos vectors;
generating a new set of block Lanczos vectors from a previous set of block Lanczos vectors;
orthogonalizing the new set of block Lanczos vectors with respect to the previous set of block Lanczos vectors;
generating a set of orthogonal factorizations corresponding to the new set of block Lanczos vectors;
generating a singular value decomposition of the new set of block Lanczos vectors;
scaling and normalizing the new set of block Lanczos vectors using the set of orthogonal factorizations and the singular value decomposition to produce a new set of scaled and normalized block Lanczos vectors;
detecting breakdowns in the new set of scaled and normalized block Lanczos vectors, and if detected, treating the breakdowns to generate a new set of starting block Lanczos vectors from the new set of scaled and normalized block Lanczos vectors; and
determining if all moments of the first expansion frequency have been used, and if all moments have not been used, generating new sets of block Lanczos vectors from the new set of starting block Lanczos vectors, and if all moments have been used, then determining if all expansion frequencies of the one or more expansion frequencies have been used, and if all of the expansion frequencies of the one or more expansion frequencies have not been used, retrieving a next expansion frequency and generating a new starting block Lanczos vectors, and if all of the expansion frequencies have been used, providing the multipoint matrix Pade approximation the electronic circuit.

14. The method of claim 13, wherein the step of detecting breakdowns further comprises the steps of:

determining that a breakdown has occurred by checking whether the new set of scaled and normalized block Lanczos vectors has a singular matrix product;
determining that the breakdown is simple and generating a first new set of random block Lanczos vectors to become a new set of starting block Lanczos vectors;
determining that the breakdown is not simple and patching the new set of scaled and normalized block Lanczos vectors with a second new set of random block Lanczos vectors to form a new set of starting block Lanczos vectors; and
orthogonalizing the new set of starting block Lanczos vectors with respect to all previously generated sets of block Lanczos vectors.

15. The method of claim 13, wherein the electronic circuit is characterized as being a linear circuit.

16. The method of claim 15, wherein the linear circuit is characterized as being an interconnect between at least two circuit elements or circuit blocks of the electronic circuit.

17. The method of claim 15, wherein the linear circuit is characterized as being a model of an analog portion of an electronic circuit.

18. The method of claim 13, wherein a set of block Lanczos vectors is characterized as being a pair of block Lanczos vectors.

19. The method of claim 13, further comprising the step of detecting if the multipoint matrix Pade approximation meets a predetermined accuracy by comparing state equations generated from the multipoint matrix Pade approximation to the state equations of the electronic circuit.

20. The method of claim 13, wherein the multipoint matrix Pade approximation is for use in circuit simulation or timing verification of the electronic circuit.

Referenced Cited
U.S. Patent Documents
5313398 May 17, 1994 Rohrer et al.
5379231 January 3, 1995 Pillage et al.
5537329 July 16, 1996 Feldmann et al.
5689685 November 18, 1997 Feldmann et al.
Other references
  • Zhaojun Bai et al,"ABLE:an Adaptive Block Lanczos Method for Non-Hermitian Eigenvalue Problems", Univ.of Kentucky, Dept. of Mathematics Res. Rpt. 95-04, May 10, 1995, pp. 1-33. Cornelius Lanczos, "An Iteration Method for the Solution of the Eigenvalue Problem of Linear Differential and Integral Operators", Journal of Res. of Nat'l. Bureau of Standards, vol.45,No.4, Oct. 1950, Res.Paper 2133, pp. 255-282. Tuyen V. Nguyen et al., "Dispersive Coupled Transmission Line Simulation using an Adaptive Block Lanczos Algorithm",Adv.Des. Tech. at Motorola,Austin, TX and Dept. of Mathematics,Univ. of Kentucky, 4 pgs. , May 5, 1996. K. Gallivan et al., "Asymptotic Waveform Evaluation via a Lanczos Method", Appl. Math. Lett.vol.7, No. 5, 1994, pp. 75-80. Peter Feldmann et al., "Efficient Linear Circuit Analysis by Pade Approximation via the Lanczos Process", IEEE Transactions on Computer-Aided Des. of Integrated Circuits and Systems, vol.14, No.5, May 1995, pp. 639-649. K.Gallivan et al., "Pade Approximation of Large-Scale Dynamic Systems With Lanczos Methods", Proceedings of the 33rd IEEE Conf.on Decision and Control,Dec.14,1994,vol.1 of 4, pp. 443-448. Mustafa Celik et al,"Pole-Zero Computation in Microwave Cir.Using Pade Approximation", IEEE Transactions on Circuits and Systems-1:Fundamental Theory and Applications, vol.42,No.1, Jan.1995, pp. 6-13. Nguyen et al. ("Dispersive Coupled Transmission Line Simulation using an Adaptive Block Lanczos Algorithm", Proceedings of the IEEE 1996 Custom Integrated Circuits Conference, May 5, 1996). Gallivan et al. ("Pade Approximation of Large-Scale Dynamic Systems with Lanczos Methods", Proceedings of the 33rd IEEE Conference on Decision and Control, Dec. 14, 1994, vol. 1, pp. 443-448. Bucker et al. ("A Parallel Version of the Unsymmetric Lanczos Algorithm and its Application to QMR", Graduiertenkolleg "Informatik und Technik", RWTH Aachen, D-52056 Aachen, Germany, Mar. 1, 1996). Gallivan et al. ("Asymptotic Waveform Evaluation via a Lanczos Method", Appl. Math. Lett., vol. 7, No. 5, 1994, pp. 75-80), Jan. 1994. Cybenko ("An Explicit Formula for Lanczos Polynomials", Linear Algebra and Its Application, vol. 88-89, pp. 99-116, Jan. 1, 1987). Su et al. ("Unsymmetric Lanczos Model Reduction and Linear State Function Observer for Flexible Structures", U.S. Government/NTIS, Sep. 1, 1991). Bandyopadhyay et al. ("Time-Domain Pade Approximation and Modal-Pade Method for Multivariable Systems", IEEE Transactions on Circuits and Systems, vol. CAS-34, No. 1, Jan. 1987). Nour-Omid et al. ("Lanczos and Arnoldi methods for the solution of convection-diffusion equations", Computer Methods in Applied Mechanics and Engineering, vol. 88, No. 1, pp. 75-95, Jun. 1, 1991). Nour-Omid ("Solving Large linearlized Systems in Mechanics", Scopus Technology, Inc., Emeryville, CA, NTIS, Mar. 1992). Simon ("The Lanczos Algorithm for Solving Symmetric Linear Systems", U.S. Government/NTIS, Apr. 20, 1982). Kim ("Block Lanczos Algorithm", U.S. Government/NTIS, Dec. 1, 1989).
Patent History
Patent number: 5920484
Type: Grant
Filed: Dec 2, 1996
Date of Patent: Jul 6, 1999
Assignee: Motorola Inc. (Austin, TX)
Inventors: Tuyen Van Nguyen (Austin, TX), Jing Li (Austin, TX)
Primary Examiner: Kevin J. Teska
Assistant Examiner: Phallaka Kik
Attorney: Daniel D. Hill
Application Number: 8/753,835
Classifications
Current U.S. Class: 364/489; 364/488; 364/490; 364/578
International Classification: G06F 1750; G06F 1716; G06F 1717;